?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Follow up to d0858bffa11, add missing REQUIRES x86
[llvm-project.git] / llvm / test / tools / llvm-mca / AMDGPU / gfx10-trans.s
blob571f1ef6c7ec0ba886b6e04ae7aa4673c98ecc00
1 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
2 # RUN: llvm-mca -mtriple=amdgcn -mcpu=gfx1010 --timeline --iterations=1 --timeline-max-cycles=0 < %s | FileCheck %s -check-prefixes=CHECK,GFX10
3 # RUN: llvm-mca -mtriple=amdgcn -mcpu=gfx1100 --timeline --iterations=1 --timeline-max-cycles=0 < %s | FileCheck %s -check-prefixes=CHECK,GFX11
4
5 v_log_f32 v0, v0
6 v_rcp_f32 v0, v0
7 v_rsq_f32 v1, v1
8 v_sqrt_f32 v2, v0
9 v_rcp_f64 v[0:1], v[0:1]
10 v_rsq_f64 v[1:2], v[1:2]
11 v_sqrt_f64 v[2:3], v[0:1]
12
13 # CHECK: Iterations: 1
14 # CHECK-NEXT: Instructions: 7
15
16 # GFX10-NEXT: Total Cycles: 94
17 # GFX11-NEXT: Total Cycles: 142
18
19 # CHECK-NEXT: Total uOps: 7
20
21 # CHECK: Dispatch Width: 1
22
23 # GFX10-NEXT: uOps Per Cycle: 0.07
24 # GFX10-NEXT: IPC: 0.07
25
26 # GFX11-NEXT: uOps Per Cycle: 0.05
27 # GFX11-NEXT: IPC: 0.05
28
29 # CHECK-NEXT: Block RThroughput: 7.0
30
31 # CHECK: Instruction Info:
32 # CHECK-NEXT: [1]: #uOps
33 # CHECK-NEXT: [2]: Latency
34 # CHECK-NEXT: [3]: RThroughput
35 # CHECK-NEXT: [4]: MayLoad
36 # CHECK-NEXT: [5]: MayStore
37 # CHECK-NEXT: [6]: HasSideEffects (U)
38
39 # CHECK: [1] [2] [3] [4] [5] [6] Instructions:
40 # CHECK-NEXT: 1 10 1.00 U v_log_f32_e32 v0, v0
41 # CHECK-NEXT: 1 10 1.00 U v_rcp_f32_e32 v0, v0
42 # CHECK-NEXT: 1 10 1.00 U v_rsq_f32_e32 v1, v1
43 # CHECK-NEXT: 1 10 1.00 U v_sqrt_f32_e32 v2, v0
44
45 # GFX10-NEXT: 1 24 1.00 U v_rcp_f64_e32 v[0:1], v[0:1]
46 # GFX10-NEXT: 1 24 1.00 U v_rsq_f64_e32 v[1:2], v[1:2]
47 # GFX10-NEXT: 1 24 1.00 U v_sqrt_f64_e32 v[2:3], v[0:1]
48
49 # GFX11-NEXT: 1 40 1.00 U v_rcp_f64_e32 v[0:1], v[0:1]
50 # GFX11-NEXT: 1 40 1.00 U v_rsq_f64_e32 v[1:2], v[1:2]
51 # GFX11-NEXT: 1 40 1.00 U v_sqrt_f64_e32 v[2:3], v[0:1]
52
53 # CHECK: Resources:
54 # CHECK-NEXT: [0] - HWBranch
55 # CHECK-NEXT: [1] - HWExport
56 # CHECK-NEXT: [2] - HWLGKM
57 # CHECK-NEXT: [3] - HWRC
58 # CHECK-NEXT: [4] - HWSALU
59 # CHECK-NEXT: [5] - HWTransVALU
60 # CHECK-NEXT: [6] - HWVALU
61 # CHECK-NEXT: [7] - HWVMEM
62
63 # CHECK: Resource pressure per iteration:
64 # CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7]
65 # CHECK-NEXT: - - - 7.00 - 7.00 3.00 -
66
67 # CHECK: Resource pressure by instruction:
68 # CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] Instructions:
69 # CHECK-NEXT: - - - 1.00 - 1.00 - - v_log_f32_e32 v0, v0
70 # CHECK-NEXT: - - - 1.00 - 1.00 - - v_rcp_f32_e32 v0, v0
71 # CHECK-NEXT: - - - 1.00 - 1.00 - - v_rsq_f32_e32 v1, v1
72 # CHECK-NEXT: - - - 1.00 - 1.00 - - v_sqrt_f32_e32 v2, v0
73 # CHECK-NEXT: - - - 1.00 - 1.00 1.00 - v_rcp_f64_e32 v[0:1], v[0:1]
74 # CHECK-NEXT: - - - 1.00 - 1.00 1.00 - v_rsq_f64_e32 v[1:2], v[1:2]
75 # CHECK-NEXT: - - - 1.00 - 1.00 1.00 - v_sqrt_f64_e32 v[2:3], v[0:1]
76
77 # CHECK: Timeline view:
78
79 # GFX10-NEXT: 0123456789 0123456789 0123456789 0123456789 0123
80 # GFX10-NEXT: Index 0123456789 0123456789 0123456789 0123456789 0123456789
81
82 # GFX11-NEXT: 0123456789 0123456789 0123456789 0123456789 0123456789 0123456789 0123456789
83 # GFX11-NEXT: Index 0123456789 0123456789 0123456789 0123456789 0123456789 0123456789 0123456789 01
84
85 # GFX10: [0,0] DeeeeeeeeeE . . . . . . . . . . . . . . . . . v_log_f32_e32 v0, v0
86 # GFX10-NEXT: [0,1] . . DeeeeeeeeeE . . . . . . . . . . . . . . . v_rcp_f32_e32 v0, v0
87 # GFX10-NEXT: [0,2] . . .DeeeeeeeeeE . . . . . . . . . . . . . . . v_rsq_f32_e32 v1, v1
88 # GFX10-NEXT: [0,3] . . . . DeeeeeeeeeE . . . . . . . . . . . . . v_sqrt_f32_e32 v2, v0
89 # GFX10-NEXT: [0,4] . . . . .DeeeeeeeeeeeeeeeeeeeeeeeE . . . . . . . . . . v_rcp_f64_e32 v[0:1], v[0:1]
90 # GFX10-NEXT: [0,5] . . . . . . . . . DeeeeeeeeeeeeeeeeeeeeeeeE. . . . . . v_rsq_f64_e32 v[1:2], v[1:2]
91 # GFX10-NEXT: [0,6] . . . . . . . . . . . . . . DeeeeeeeeeeeeeeeeeeeeeeeE v_sqrt_f64_e32 v[2:3], v[0:1]
92
93 # GFX11: [0,0] DeeeeeeeeeE . . . . . . . . . . . . . . . . . . . . . . . . . .. v_log_f32_e32 v0, v0
94 # GFX11-NEXT: [0,1] . . DeeeeeeeeeE . . . . . . . . . . . . . . . . . . . . . . . .. v_rcp_f32_e32 v0, v0
95 # GFX11-NEXT: [0,2] . . .DeeeeeeeeeE . . . . . . . . . . . . . . . . . . . . . . . .. v_rsq_f32_e32 v1, v1
96 # GFX11-NEXT: [0,3] . . . . DeeeeeeeeeE . . . . . . . . . . . . . . . . . . . . . .. v_sqrt_f32_e32 v2, v0
97 # GFX11-NEXT: [0,4] . . . . .DeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeE . . . . . . . . . . . . . . . .. v_rcp_f64_e32 v[0:1], v[0:1]
98 # GFX11-NEXT: [0,5] . . . . . . . . . . . . .DeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeE . . . . . . . .. v_rsq_f64_e32 v[1:2], v[1:2]
99 # GFX11-NEXT: [0,6] . . . . . . . . . . . . . . . . . . . . .DeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeE v_sqrt_f64_e32 v[2:3], v[0:1]
100
101 # CHECK: Average Wait times (based on the timeline view):
102 # CHECK-NEXT: [0]: Executions
103 # CHECK-NEXT: [1]: Average time spent waiting in a scheduler's queue
104 # CHECK-NEXT: [2]: Average time spent waiting in a scheduler's queue while ready
105 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
106
107 # CHECK: [0] [1] [2] [3]
108 # CHECK-NEXT: 0. 1 0.0 0.0 0.0 v_log_f32_e32 v0, v0
109 # CHECK-NEXT: 1. 1 0.0 0.0 0.0 v_rcp_f32_e32 v0, v0
110 # CHECK-NEXT: 2. 1 0.0 0.0 0.0 v_rsq_f32_e32 v1, v1
111 # CHECK-NEXT: 3. 1 0.0 0.0 0.0 v_sqrt_f32_e32 v2, v0
112 # CHECK-NEXT: 4. 1 0.0 0.0 0.0 v_rcp_f64_e32 v[0:1], v[0:1]
113 # CHECK-NEXT: 5. 1 0.0 0.0 0.0 v_rsq_f64_e32 v[1:2], v[1:2]
114 # CHECK-NEXT: 6. 1 0.0 0.0 0.0 v_sqrt_f64_e32 v[2:3], v[0:1]
115 # CHECK-NEXT: 1 0.0 0.0 0.0 <total>
LLVM monorepo